Can you explain to us why?

Yes, and thank you.

Um, there is sort of two layers to this.

The first is that oftentimes in the discourse, people talk about artificial intelligence as if it were one thing, rather than a collection of actually quite diverse technologies that are each built with whatever different algorithm and whatever different training data and hopefully evaluated on appropriate data as well.

So the systems that have to do with the, uh, predicting the shape of a folded protein are not the same piece of software as, um, the automatic player you might encounter in a video game, which is not the same piece of software as your automatic transcription system or machine translation or chatbot.

And I think that the discourse gets clearer if we actually specify what technology we're talking about.

That's the first problem.

W- why, why should we be taking their words with caution?

What do you think about that assertion?

It is a misdirection, and it's also part of the hype.

So AI safety is the name for protecting people from, in quotes, AI systems.

And it's not protecting people who are arrested after a false match in a facial recognition system.

It's not protecting people who are working under horrible conditions cleaning up the outputs of these systems with no mental health supports.

for $2 an hour.

It's not about protecting the residents of Box Town near Nashville who are breathing in all the pollutants off of the methane generators that Elon Musk is using to power his data center, right? These are real safety issues.

I think it would be really interesting- For you to explain, when you say you're not interested in the project of artificial intelligence, can you really lay out what that project is? What do you understand the project to

Um, but what is your current stance on what people currently talk of as AI, and how did you come to that view of this technology?

Yeah.

So I think the very first thing is that the phrase is problematic on many levels.

It is anthropomorphizing to talk about these systems as if they have intelligence.

Dig one layer deeper, the whole notion of intelligence is just like a, a racist cesspool.

We don't need to go there.

The idea that you can line people up, um, according to one particular property, and that property is sort of this key idea of general intelligence comes to us straight from eugenics and race science.

Mm-hmm.

Right.

Yeah.

Um, but if we want to distinguish sort of old-fashioned web search from, uh, the so-called AI search or chat interface, um, the big difference is that, yes, the, the way you interact with it is instead of putting in some keywords that you hope will match the documents you're looking for, you phrase the question the way you might ask a person who could understand you, and that helps create the illusion that the system you are interacting with does understand you.

But fundamentally, these systems are just set up to repeatedly answer the question: what is a likely next word? And what is a likely next word, given that question as input and, uh, the various documents that the web search returned, produces something that looks like an answer and is very easy to accept as an answer, but we are then cut off from really important sense-making work where we understand where information is coming from, why somebody may have put those particular pieces of information together, and how to situate all of that in our understanding of the information landscape.

But boy, sometimes it really hits it out of the park.

Yeah.

As you might guess, I have some objections to that, and I sure hope that you don't do it for this segment.

Um, and by the time listeners are hearing this, they will have heard whatever it was, right? [laughs] Um, but I...